@article { author = {Ansari Ghojghar, Mohammad and Bazrafshan, Javad and Araghinejad, Shahab and Parsi, Ehsan and Soltani, Shakour}, title = {Evaluation of the performance of the support-wavelet vector machine hybrid model in predicting dust storms (Case study: Sistan and Baluchestan province)}, journal = {Environmental Management Hazards}, volume = {7}, number = {4}, pages = {331-351}, year = {2020}, publisher = {}, issn = {2423-415X}, eissn = {2423-4168}, doi = {10.22059/jhsci.2021.314582.614}, abstract = {Introduction In recent years, the use of combining meta-models with optimization algorithms to predict hydrological and meteorological variables has increased, some of which are mentioned below. The results of research on drought prediction using genetic algorithm and hybrid neural-wave network model showed that the application of the combined method in comparison with the combination of genetic algorithm and neural network provides desirable results [5]. The study of the performance of hybrid models of artificial neural network and support vector machine in estimating the discharge of Zarrinehrood River, located in Iran, showed that the hybrid model of artificial neural network has better accuracy than support vector machine [7]. Long-term rainfall in Anzali over a period of 67 years was assessed by the wavelet numerical model - adaptive neural fuzzy inference system (W-ANFIS). The results of this study showed that the W-ANFIS model with the values ​​of correlation coefficient, dispersion index and cluster instability equal to 0.962, 0.258 and 0.899, respectively, has a good ability to simulate rainfall phenomenon [6]. In this study, SVM meta-model and its combined type with artificial plants (AF) and wavelet (W) algorithms will be evaluated in order to predict the FDSD index in five synoptic stations of Sistan and Baluchestan province. Therefore, this study can be a new approach in how to use intelligent hybrid methods to predict the FDSD index in the study area.   Materials and methods In this study, two hybrid methods under the headings of support vector-wavelet (W-SVM) and support vector-artificial plants algorithm (AF-SVM) with individual model of support vector machine (SVM) to predict the frequency of days with Dust storm (FDSD) in five synoptic stations of Sistan and Baluchestan province (Zabol, Zahedan, Iranshahr, Khash and Saravan) were compared with the long-term statistical population of 40 years (1980-1920) on a seasonal scale. For this purpose, horizontal vision power data and WMO codes were used. Observations of meteorological phenomena are recorded every three hours, eight times a day. In these observations, visual phenomena of the weather are defined in 100 codes (00-99) according to the instructions of the World Meteorological Organization, of which 11 codes are generally used to record and report dust phenomena in different meteorological stations [8]. In this study, a horizontal visibility factor was used for all dust meteorological codes to detect dust storms. After selecting the stations and reviewing the data over a period of 40 years (1980-1920), the number of days with dust storm (FDSD) for the five meteorological stations studied in Sistan and Baluchestan province was calculated using horizontal visibility data and Meteorological Organization codes. While, meteorological stations, latitude and longitude, altitude, average FDSD index on a seasonal scale, and the number of dust days can be seen in ascending order. Results and discussion The results of FDSD index forecast indicate the good performance of both AF-SVM and W-SVM hybrid methods in all studied stations (Zabol, Zahedan, Iranshahr, Khash and Saravan). Another point is the poor performance of the individual backup vector machine model compared to both hybrid methods. The hybrid support vector-wavelet model in all studied stations has better accuracy and overlap than other studied models due to the optimization of model parameters by the wavelet algorithm. On the other hand, the support vector machine model-artificial plants algorithm has shown a good performance after the hybrid model of the support vector machine-wavelet. On the other hand, according to the root mean square error and Nash Sutcliffe coefficient, the hybrid backup-wavelet machine hybrid model, again, showed less error and higher accuracy in all selected stations. Mean Absolute Error in two synoptic stations of Saravan and Khash, which have the lowest number of dusty days, the hybrid model of the support vector machine-artificial plants algorithm performed better than the hybrid model of the support vector-wavelet model. However, in other studied stations, such as good fit criteria of the previous one, the hybrid backup-wavelet machine hybrid model performed as the best. In addition, the hybrid model of the support vector machine-artificial plants algorithm has a good accuracy in predicting the intermediate and maximum values, and as we move towards the minimum values, it increases the accuracy and efficiency of the hybrid model of the support vector machine-artificial plants algorithm. The selection of more complex models is the optimal predictive model in the studied stations; in order to predict the FDSD index in all stations, models 3 and 4 (with three and four steps delay) were used, which can be due to the impact of particles left over from previous storms and seasons. The past has looked to the formation of dust storms next season. The results of this section are consistent with studies conducted in this field [1, 2, 3, 4 and 7]. The performance of all methods for predicting the FDSD index is directly related to the increase in the number of days associated with dust storms. The results of this section are in line with studies conducted in this field [4]. Conclusion The results of this study showed that the use of a combined support vector-wavelet model could be very effective in predicting the frequency of days with dust storms. In addition, in all the methods used to predict the FDSD index, the model that used three or four steps of delay in forecasting was the best predictive model, which can be due to the effect of particles. The remnants of previous storms and the previous season(s) sought to form next season's dust storms. Considering that the decision to control dust storms and implement management strategies in many critical areas of the country depends on the accurate estimation of dust storms; therefore, using the proposed hybrid model to predict the FDSD index, can be used as an appropriate tool in management decisions. Undoubtedly, in order to validate the results of this study, more research should be done on the application of hybrid meta-models in modeling and temporal-spatial prediction of dust storms in areas affected by this phenomenon. It is also suggested to use hybrid models of support vector machine with new optimization algorithms such as skiing, chicken swarming, cat swarming, creative rifleman, etc. and compare with the results of the proposed model.}, keywords = {Artificial Flora,frequency,Dust Storm,Support Vector Machines,Sistan and Baluchestan Province}, title_fa = {بررسی عملکرد مدل هیبریدی ماشین بردار پشتیبان- موجک درپیش‌بینی توفان‌های گردوغبار (مطالعۀ موردی: استان سیستان و بلوچستان)}, abstract_fa = {به‌منظور کنترل و مدیریت صحیح توفان‌های گردوغبار، آگاهی از تغییرات زمانی این پدیده و لزوم پیش‌بینی و مدل‌سازی آن ضروری است. در این پژوهش به‌منظور پیش‌بینی متغیر فراوانی روزهای همراه با توفان گردوغبار (FDSD)، نتایج دو روش هیبریدی با نام ماشین بردار پشتیبان- موجک (W-SVM) و ماشین بردار پشتیبان- الگوریتم گیاهان مصنوعی (AF-SVM) به‌همراه مدل انفرادی ماشین بردار پشتیبان (SVM)، مقایسه شد. بدین منظور از داده‌های ساعتی گردوغبار و کدهای سازمان جهانی هواشناسی در مقیاس فصلی با طول دورۀ آماری چهل‌ساله (2018-1980) در پنج ایستگاه سینوپتیک منتخب استان سیستان و بلوچستان استفاده شد. معیارهای ضریب تبیین، ریشۀ میانگین مربعات خطا، میانگین قدرمطلق خطا و ضریب نش‌ساتکلیف برای ارزیابی و مقایسۀ مدل‌ها، استفاده شد. نتایج در مرحلۀ آموزش و آزمایش نشان داد که ساختارهای ترکیبی استفاده‌شده، نتایج قابل قبولی در مدل‌سازی شاخص FDSD ارائه می‌کنند. مدل هیبریدی ماشین بردار پشتیبان- موجک با ضریب همبستگی (984/0-911/0R2=)، ریشۀ میانگین مربعات خطا (day 314/0-397/0RMSE=)، میانگین قدر مطلق خطا (day 335/0-236/0MAE=) و ضریب نش‌ساتکلیف (965/0-924/0NS=)، عملکرد بهتری نسبت به دیگر مدل‌های استفاده‌شده در پیش‌بینی شاخص FDSD داشته است. نتایج این تحقیق می‌تواند در مدیریت پیامدهای ناشی از توفان‌های گردوغبار و برنامه‌های مقابله با بیابان‌زایی در مناطق تحت مطالعه مؤثر واقع شود.}, keywords_fa = {Artificial Flora,frequency,Dust Storm,Support Vector Machines,Sistan and Baluchestan Province}, url = {https://jhsci.ut.ac.ir/article_80413.html}, eprint = {https://jhsci.ut.ac.ir/article_80413_3ebd1678e5512747ae0ebe843c2e62e2.pdf} }